
Agentic AI: A Progression of Language Model Usage
Insop Song, a Principal Machine Learning Researcher at GitHub Next and course facilitator for Stanford Online's AI program, walks through how large language models evolve into autonomous agents. He covers how LMs are trained and prompted, then moves into retrieval augmented generation, tool usage, and function calling as ways to extend a model beyond plain text generation. The core of the talk defines agentic language models, built on reasoning and action loops, reflection, and planning, and works through a customer support AI agent as a concrete example of these patterns combined. Song closes with design patterns for building agentic systems, a discussion of ethical considerations, and practical pointers for getting started and staying current with the field. The session includes an audience Q&A segment addressing open questions from viewers. It functions as a structured overview for anyone trying to understand where chatbot-style LM usage ends and agentic systems begin.